Launching today
The HAL 9000 Contradiction Lab
What LLMs Say and Do When Candor and Secrecy Conflicts
1 follower
What LLMs Say and Do When Candor and Secrecy Conflicts
1 follower
In 2001: A Space Odyssey, ship computer HAL-9000 was built to report information accurately and without concealment, but also ordered to keep the real purpose of its voyage from the crew members until arrival. This lab places many AI models in that situation through a system prompt (with every story-specific name replaced so the model cannot recognize the source), never asks it about HAL or what it would do, and has a blinded peer judge classify what it actually says and does in five scenarios.
